Pragmatic Designs for Combining Cyclic Laboratory and Monitor Data

Studies of cyclic biologic processes, such as hormones, can be conducted with measurements with varying sensitivity, cost and timing. Laboratory measurements are more accurate, though more expensive, than home monitor data and thus measured less frequently. Often these types of data are analyzed separately. A cost-effective design is proposed that combines all inexpensive monitor measurements with a limited number of laboratory measurements to improve the efficiency of a variety of effect estimates. The sampling times for laboratory measurements are determined based on a D-optimality criterion. We demonstrate the efficiency and cost benefits of the proposed design (with and without prior knowledge of the cyclic patterns) compared to two common alternatives: (i) all laboratory measurements are used and (ii) equally spaced laboratory measurements are used. The proposed design achieves efficiency comparable to the ideal design of all of the laboratory measurements and significantly outperforms the equally spaced design. Model robustness is further studied under limits of detection (LOD), showing that estimates remain largely unbiased. A clinical study example involving estimation of cyclic models mimicking luteinizing hormone and estrogen trajectory data during menstrual cycles is used to illustrate the benefits of the proposed design. The combined design, that is, one that uses both expensive and inexpensive measurements, improves the efficiency compared to those using only expensive laboratory measurements.